Blind Image Quality Assessment via Convolutional Neural Network

Abstract

This paper proposed a novel blind image quality assessment method that is created by training a convolutional neural network to learn discriminant features of image quality and fitting the features with a support vector regression to get an evaluation score. The pooling procedure is help to reduce the feature dimension and improve computation efficiency. The proposed method does not need any hand-crafted features contrast with most previous BIQA methods. It achieves better performance than previous BIQA methods on LIVE database. The experimental results show that the proposed method has good consistency, robustness and efficiency.